LaHuis David M, Blackmore Caitlin E, Ammons Gage M
Wright State University, Dayton, OH, USA.
Aon Hewitt, Lincolnshire, IL, USA.
Appl Psychol Meas. 2025 Jan 27:01466216251316278. doi: 10.1177/01466216251316278.
This study compared maximum a posteriori (MAP), expected a posteriori (EAP), and Markov Chain Monte Carlo (MCMC) approaches to computing person scores from the Multi-Unidimensional Pairwise Preference Model. The MCMC approach used the No-U-Turn sampling (NUTS). Results suggested the EAP with fully crossed quadrature and the NUTS outperformed the others when there were fewer dimensions. In addition, the NUTS produced the most accurate estimates in larger dimension conditions. The number of items per dimension had the largest effect on person parameter recovery.
本研究比较了最大后验概率（MAP）、期望后验概率（EAP）和马尔可夫链蒙特卡罗（MCMC）方法，以从多维度成对偏好模型计算个人得分。MCMC方法使用了无回转采样（NUTS）。结果表明，在维度较少时，具有完全交叉求积的EAP和NUTS方法优于其他方法。此外，在维度较大的条件下，NUTS方法产生的估计最为准确。每个维度的项目数量对个人参数恢复的影响最大。
